Measurement system

Abstract

The purpose of the present invention is to provide a measurement system capable of preparing a measurement parameter used for measuring a structure of a complicated pattern at a low cost. A measurement system according to the present invention infers a measurement region from an image of a sample, extracts a feature amount of the inferred measurement region, and uses a measurement parameter corresponding to the feature amount to measure the sample using the measurement parameter corresponding to the feature of the measurement region (see fig. 1).

Description

Measurement System

[0001] The present invention relates to a measurement system for measuring a structure.

[0002] As semiconductors become more miniaturized and multi-layered, there is a demand for measuring complex patterns. Measuring complex patterns requires the development of a pattern-specific measurement method, which requires time and human resources, so there is a demand for technology that enables measurement at low cost. In conventional measurements, measurement points are calculated using measurement parameters prepared in advance.

[0003] In the patent document 1 listed below, the model and parameters of the material and shape of the sample to be measured are registered in advance in a library, the waveform of the captured image is matched with the library, and the measurement is performed using the matched model and parameters.

[0004] JP 2009-198339 A

[0005] As patterns become more complex, the variation in features within the measurement area increases, and the optimal measurement parameters differ for each measurement location. Therefore, in a technology like Patent Document 1, in which measurement parameters are registered in a library in advance, in order to accommodate the measurement of increasingly complex patterns, it is necessary to build a library in advance for a wide variety of sample shapes and parameters, which is costly.

[0006] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a measurement system that can prepare measurement parameters used to measure complex pattern structures at low cost.

[0007] The measurement system of the present invention infers a measurement area from an image of a sample, extracts features of the inferred measurement area, and measures the sample using the measurement parameters corresponding to the features by using the measurement parameters corresponding to the features of the measurement area.

[0008] According to the measurement system of the present invention, it is possible to prepare measurement parameters used for measuring the structure of a complex pattern at low cost. Other objects, configurations, advantages, etc. of the present invention will become apparent from the description of the following embodiments.

[0009] 10 shows an example of the configuration of a measurement system 100 according to embodiment 1. FIG. 11 is a flowchart showing an example of a procedure for preparing a measurement area inference model D102, a feature amount classification model D103, and measurement parameters D104. FIG. 12 shows an example of a method for training the measurement area inference model D102 in S201. FIG. 13 shows an example of a method for training the feature amount classification model D103 in S202. FIG. 14 shows an example of a method for searching for measurement parameters D104 for each classification class in S204. FIG. 15 shows an example of the configuration of a measurement parameter search unit 503. FIG. 16 shows an example of a procedure for the measurement processing unit 104 to calculate measurement values. FIG. 17 shows an example of a procedure for the measurement processing unit 104 to calculate measurement points. FIG. 18 is a diagram explaining an example of index values ​​using a scatter plot of true values ​​and measured values. FIG. 19 shows an example when the same value is used for the predetermined ratio in the threshold method in various measurement areas. FIG. 19 shows an example in which the rate is set to 0.8 only in (3) of FIG. 10. FIG. 19 is a diagram showing the configuration of a measurement system 100 according to embodiment 2. 1 shows an example of a specific method for the outlier reporting unit 1201 to report an outlier. 2 shows an example of a GUI provided by the measurement system 100. 3 shows an example of a GUI provided by the measurement system 100.

[0010] <First Embodiment> Fig. 1 shows an example of the configuration of a measurement system 100 according to a first embodiment of the present invention. The measurement system 100 is a system that measures the structure of a sample. The measurement system 100 receives an input image D101, a measurement area inference model D102, a feature amount classification model D103, and measurement parameters D104. The measurement system 100 includes a measurement area inference unit 101, a feature amount classification unit 102, a parameter setting unit 103, and a measurement processing unit 104.

[0011] The measurement area inference unit 101 uses an input image D101 and a measurement area inference model D102 to extract feature amounts D105 related to the measurement area of ​​the input image D101, and infers a measurement area D106 based on the extracted feature amounts D105. The reason for inferring the measurement area D106 will be described later.

[0012] The feature amount classification unit 102 classifies the feature amounts into classes using the feature amounts D105 and the feature amount classification model D103. The input feature amounts indicate the features of the measurement region, and the classification results D107 are classified according to the features of the measurement region.

[0013] The parameter setting unit 103 receives the classification result D107 and the measurement parameters D104 as input, and sets parameters required for measurement processing for each classification class.

[0014] The measurement processing unit 104 performs measurement processing on the measurement area D106 using parameters set for each measurement area D106 and each classification class, and outputs the measurement result D108.

[0015] The input image D101 is an image that shows an object to be measured, such as its position or dimensions. The input image D101 may be, for example, an inspection image of a semiconductor captured by a scanning electron microscope (SEM), a medical image captured by a computerized tomography (CT), an inspection image of a component, or an image of a public road. The input image and the object to be measured are not limited. There may be one or more objects to be measured in a single image, and the number is not limited. In addition, there may be not just one but multiple input images. The position and size of the object to be measured vary from image to image and from object to object, so it is necessary to set a measurement area for each object to be measured.

[0016] The measurement area inference model D102 is a model that infers a measurement area for searching for a measurement location in the input image D101. The measurement area inference model D102 is constructed, for example, by a CNN (Convolution Neural Network). The measurement area inference model D102 may use, for example, object detection AI (artificial intelligence) that detects the position of an object from an image, or may be a model that infers a class for each pixel using semantic segmentation AI and detects the position of a specific shape as the position of the object. Alternatively, a model that does not use a CNN but calculates feature points from feature amounts that are invariant to the position and size of an object in an image and detects feature points of a measurement target may be used. Other models may also be used.

[0017] The measurement area inference unit 101 infers a measurement area D106 in the input image D101 using a measurement area inference model D102. At this time, a feature amount D105 extracted from the input image D101 when inferring the measurement area D106 is also output. The feature amount D105 is a feature amount that serves as a basis when the measurement area inference model D102 infers the measurement area D106. The feature amount D105 may be a vector in an intermediate layer of a CNN if, for example, an object detection AI or semantic segmentation AI configured with a CNN is used as the measurement area inference model D102, or may be a feature amount that is invariant to the position or size of an object in the image if the measurement area D106 is inferred by calculating feature points of the image. Other calculation methods or feature amounts may be used as long as they are feature amounts that serve as a basis when inferring the measurement area D106. If the measurement area inference model D102 can correctly infer the measurement area D106 for an input image having a variety of features, the extracted feature amounts indicate different vectors and values ​​for each feature of the measurement area D106.

[0018] The feature classification model D103 is a model that classifies classes based on the feature D105. For example, it may be a classification model configured by a neural network (NN), or a model such as Kmeans that compares the distances between feature values ​​and clusters feature values ​​with small distances, and the configuration of the model is not limited.

[0019] The feature amount classification unit 102 inputs the feature amount D105 to the feature amount classification model D103, and outputs a classification result D107. As described above, the feature amount D105 indicates a different characteristic for each feature of the measurement area D106, and therefore the measurement area D106 can be classified based on the characteristics of the measurement area D106.

[0020] The measurement parameters D104 correspond to the classes classified by the feature amount classification model D103, and are measurement parameters for each classification class used by the measurement processing unit 104. The measurement parameters D104 are appropriate parameters that improve the measurement accuracy for each classification class.

[0021] The parameter setting unit 103 sets appropriate parameters for each measurement region D106 using the classification result D107 and the measurement parameters D104. Since the classification result D107 is classified based on the characteristics of the measurement region D106, it is possible to set appropriate parameters for each measurement region D106.

[0022] The measurement processing unit 104 performs measurement processing using the parameters set by the parameter setting unit 103. For example, the measurement processing may be a method of detecting edges of a measurement target using a threshold method in the case of a semiconductor image and calculating measurement values ​​from the edges, or another method. In this case, parameters may be prepared in advance depending on the measurement processing, and appropriate parameters may differ for each feature of the measurement region. Even in this case, appropriate parameters are set by the parameter setting unit 103 for each feature of the measurement region D106, so appropriate measurement processing is possible.

[0023] 2 is a flowchart showing an example of a procedure for preparing a measurement area inference model D102, a feature quantity classification model D103, and measurement parameters D104. This flowchart may be performed, for example, by one of the functional units included in the measurement system 100, or by a computer separate from the measurement system 100. For convenience of description below, it is assumed that the measurement system 100 performs each step of the flowchart. Each step in FIG. 2 will be described below.

[0024] S201: The measurement system 100 uses the training data D201 to train and construct a measurement area inference model D102. The contents of the training data D201 and the procedure for training the measurement area inference model D102 will be described later.

[0025] S202: The measurement system 100 extracts features from the training data D202 using the trained measurement area inference model D102, and trains a feature classification model D103 based on the extracted features. The contents of the training data D202 and the procedure for training the feature classification model D103 will be described later.

[0026] S203: In order to set appropriate classification class parameters, the measurement system 100 first extracts feature quantities from the parameter search data D203 using the trained measurement area inference model D102. The measurement system 100 classifies the feature quantities using the feature quantity classification model D103. By classifying the feature quantities, classification classes D204 representing each classification can be obtained. The contents of the parameter search data D203 and the procedure for obtaining the classification classes D204 will be described later.

[0027] S204: The measurement system 100 uses the classification class D204 and the index value indicating the measurement accuracy to search for classification class parameters so as to improve the index value, thereby obtaining the measurement parameter D104. The index value can be calculated according to an index value calculation method D205. The content of the index value and the procedure for searching the measurement parameter D104 will be described later.

[0028] The training data D201 is data that enables the measurement area inference model D102 to correctly infer the measurement area D106, and is, for example, a pre-annotated image and measurement area. The format of the training data D201 differs for each measurement area inference model D102.

[0029] The training data D202 is a group of images used to train the feature classification model D103. The training data D202 may include data indicating classes as supervised information in the case of supervised learning, or may consist of images only in the case of unsupervised learning such as KMeans, and the data format differs depending on the learning method.

[0030] The parameter search data D203 is data used to search for appropriate parameters for each classification class. The parameter search data D203 is in a format corresponding to a parameter search method, and is composed of, for example, only an image.

[0031] The image data included in the learning data D201, the learning data D202, and the parameter search data D203 may overlap, or may be different image data.

[0032] The index calculated by the index calculation method D205 is, for example, an index indicating measurement accuracy. Examples of the index include the error between a true measurement value and a measurement value calculated based on inferred information, and the correlation coefficient between a true value calculated using multiple data and an inferred value.

[0033] If the feature classification model D103 is a model constructed by unsupervised learning, the only data other than images among the data (D201, D202, D203, D205) used in Figure 2 are the measurement area and index values ​​of the learning data D201, so the cost required to prepare this data is small.

[0034] 3 shows an example of a method for training the measurement area inference model D102 in S201. If the measurement area inference model D102 is an object detection AI using CNN, the training data D201 is an image D301 and a measurement area D302. The measurement area D302 is a measurement area for calculating measurement values ​​of the image D301. The measurement area D302 can be prepared, for example, by a user manually setting the measurement area D302, actually performing measurement processing, and confirming that the processing results are correct. Other methods may also be used for preparation.

[0035] A measurement area inference unit 301 included in the measurement system 100 obtains a measurement area D304 by inferring a measurement area of ​​an image D301 using a measurement area inference model D102. A measurement area inference model update unit 302 included in the measurement system 100 compares the measurement area D304 with the measurement area D302 in the learning data D201, and updates the weight parameters of the CNN of the measurement area inference model D102 so as to reduce the difference between the two.

[0036] Possible methods for the measurement area inference model update unit 302 to calculate the difference between the measurement area D302 and the measurement area D304 include, but are not limited to, a method of comparing the center of gravity positions of the areas, a method using GIoU (Generalized Intersection over Union), a method combining these, etc. Furthermore, possible methods for updating the weight parameters of the CNN of the measurement area inference model D102 include, but are not limited to, SGD (Stochastic Gradient Descent).

[0037] Through the above procedure, the measurement area (i.e., D304) inferred by the measurement area inference model D102 approaches the measurement area D302 of the training data D201, enabling correct inference. In order to improve the generalization performance of the measurement area inference model D102, it is also possible to perform training by padding the data by inverting or enlarging or reducing the images in the training data D201. The preprocessing of the training data D201 is not limited to this.

[0038] FIG. 4 shows an example of a method for training the feature classification model D103 in S202. The feature extraction unit 401 included in the measurement system 100 extracts feature values ​​D401 from the training data D202 using the measurement area inference model D102. The feature classification model training unit 402 included in the measurement system 100 trains the feature classification model D103 using the feature values ​​D401. The training method varies depending on the format of the feature classification model D103. For example, if the feature classification model D103 is a model using a neural network (NN), a method may be used in which a teacher class is compared with an inferred class and the weight parameters of the NN are updated. Alternatively, if a method such as KMeans is used, a corresponding training method may be used; the training method is not limited. If the classification model can be trained without a teacher, such as KMeans, the training data D202 may consist of only images, and the cost of advance preparation is low. Furthermore, in order to improve generalization performance, the feature classification model learning unit 402 may perform preprocessing such as adding noise to the feature D 401. The method of preprocessing is not limited.

[0039] 5 shows an example of a method for searching for measurement parameters D104 for each classification class in S204. A feature extraction unit 501 included in the measurement system 100 extracts a measurement area D501 and its feature amounts D502 from the parameter search data D203 using a measurement area inference model D102. A feature classification unit 102 classifies the feature amounts D502 using a feature amount classification model D103. By classifying the feature amounts D502, a classification class D503 representing each classification is obtained. A measurement parameter search unit 503 included in the measurement system 100 searches for measurement parameters D104 using the classification class D503 and the measurement area D501.

[0040] 6 shows an example of the configuration of the measurement parameter search unit 503. First, the measurement processing unit 104 performs measurement processing using a measurement region D501 and a classification class D503. The measurement processing method differs depending on the measurement target, and for example, in the case of an inspection image of a semiconductor, a method of calculating measurement values ​​using a threshold method is conceivable.

[0041] The index value calculation unit 5032 included in the measurement parameter search unit 503 calculates the index value using the index value calculation method D205. Examples of index values ​​include the average value of errors between the measured values ​​and the true values ​​when multiple images are measured, and the correlation coefficient between the true values ​​and the measured values. This index value may be freely determined depending on the purpose of the measurement. The parameter setting unit 5033 included in the measurement parameter search unit 503 sets the parameter D601 based on the index value. As an example of a method for setting the parameter D601, if the function for calculating the index value is differentiable, parameters that improve the index value may be analytically calculated. Alternatively, if the function for calculating the index value is not differentiable, parameters may be searched for using a technique called black-box optimization such as Bayesian optimization, or other methods may be used. When Bayesian optimization or the like is used, it is possible to calculate parameters that improve the index value as much as possible by searching multiple times.

[0042] 7 shows an example of the procedure by which the measurement processing unit 104 calculates measurement values. In this example, the line width of a vertical bar as shown in D701 is calculated. To calculate the line width, measurement areas, which are areas for searching for measurement points, are placed around the left and right ends of the vertical bar. Next, measurement points within the measurement areas are calculated to obtain measurement points as shown in D702. Next, the line width is calculated by calculating the distance between the measurement points.

[0043] 8 shows an example of the procedure by which the measurement processing unit 104 calculates measurement points in FIG. 7. First, a line profile is created from the measurement area indicated by the dotted line in D701. For example, the line profile is created by taking the distribution of luminance values ​​along a direction where the luminance gradient is large and the edge is easy to calculate (the horizontal direction in this example). If the luminance values ​​contain a lot of noise at this time, smoothing processing may be performed. Possible smoothing methods include applying an averaging filter in the vertical or horizontal direction or applying a Hamming window, and the method is not limited. The parameters of the smoothing method must be determined in advance depending on the pattern.

[0044] Next, measurement points are determined based on the line profile. One example of a determination method is the threshold method. The threshold method is a technique in which a brightness value having a predetermined ratio between the maximum and minimum values ​​of the line profile is set as the threshold (Equation 1), and the coordinates having a value corresponding to the threshold are set as measurement points, thereby making it possible to determine the measurement points. The predetermined ratio is a parameter determined in advance. In addition to the threshold method, there is also a method in which the coordinates at which the slope of the line profile is maximum are set as measurement points, and the method for determining the measurement points is not limited. Threshold = min + (max - min) x rate (Equation 1)

[0045] The strength of smoothing when creating a line profile and the rate when calculating the threshold are parameters that are determined in advance, but it is possible to automatically search for a method that will most improve the index value using the method in Fig. 6. It is also possible to prepare multiple methods for smoothing the line profile and methods for determining measurement points, and search for a method that will most improve the index value.

[0046] FIG. 9 is a diagram illustrating an example of index values ​​using a scatter plot of true values ​​and measured values. In FIG. 9, the true values ​​are plotted on the horizontal axis and the measured values ​​are plotted on the vertical axis. In this scatter plot, when a linear approximation is performed in the form of y = ax + b, a is called the slope and is one of the index values. The slope indicates the average ratio of the change in the measured value to the change in the magnitude of the true value, with the closer it is to 1, the better the accuracy. The offset b is also called the offset and is also one of the index values, with the closer it is to 0, the better the accuracy. The coefficient of determination for this approximated line is also one of the index values. The coefficient of determination indicates the goodness of fit of the approximated line, with the closer it is to 1, the better the accuracy. The average and variance of the error between the true value and the measured value are also index values. Other values ​​may be used as index values. The parameter setting unit 5033 searches for parameters D601 so as to improve index values ​​in line with the user's measurement objectives. The parameters D601 may also be searched for using a technique called multi-objective optimization, which improves multiple index values. There is no limit to the number of index values ​​that can be improved.

[0047] Figure 10 shows an example where the same value for the predetermined ratio in the threshold method is used for various measurement regions. In Figure 10, there is one vertically elongated pattern with a slightly brighter luminance value and three horizontally elongated patterns with even brighter luminance values ​​against a dark background. The horizontally elongated patterns are located at various positions relative to the vertically elongated pattern. The measurement target is the left edge of the horizontal side of the horizontally elongated pattern, and the measurement regions are positioned around it as shown in (1), (2), and (3). The resulting line profiles are shown in the center diagram. In this case, if the predetermined ratios in the threshold method shown in Figure 8 are all set to 0.5, the measurement points are determined as shown in the diagram on the right. Measurement points are determined in approximately the correct positions for (1) and (2), but for (3), the measurement point is not determined at the left edge of the horizontally elongated pattern. This shows that a rate of 0.5 is inappropriate when observing the profile of (3).

[0048] Figure 11 shows an example in which the rate for only (3) in Figure 10 is set to 0.8. Checking the diagram on the right shows that the measurement point for (3) is positioned at the left end of the horizontally long pattern. In the measurement system 100, (1), (2), and (3) each have different characteristics, so they are classified into different classification classes and can be measured using different rates. Furthermore, the rate value can be automatically searched for using the method shown in Figure 2, enabling appropriate measurements with minimal preparation costs.

[0049] <Embodiment 2> In embodiment 2 of the present invention, an example of a configuration will be described that uses the features calculated in embodiment 1 to detect a measurement area having abnormal features using a feature classification model and has a function of issuing an alert as an outlier.

[0050] 12 is a configuration diagram of a measurement system 100 according to the second embodiment. In the second embodiment, in addition to the configuration described in the first embodiment, an outlier reporting unit 1201 is provided. The outlier reporting unit 1201 detects a measurement area having abnormal characteristics using the feature D105 and the feature classification model D103, and reports the measurement area as an outlier. The other configurations are the same as those of the first embodiment.

[0051] FIG. 13 shows an example of a specific method by which the outlier reporting unit 1201 reports an outlier. The feature classification model D103 is configured to calculate the center position of each cluster using a method such as KMeans, and each measurement area belongs to the cluster whose feature is closest to the center position of the cluster. A threshold is set for the distance from the center position of the cluster, and a measurement area that is farther away from any cluster center than the threshold can be considered an outlier. This threshold can be automatically set, for example, by setting it to three times the data variance for each cluster during training of the feature classification model D103. The outlier reporting unit 1201 can detect outliers using the procedure described above.

[0052] Third Embodiment In a third embodiment of the present invention, an example of a GUI (Graphical User Interface) will be described, which displays the measurement areas, measurement points, and classification classes to which the measurement areas belong, calculated in the first and second embodiments, and displays classification class parameter information. The other configurations are the same as those in the first and second embodiments.

[0053] 14 shows an example of a GUI provided by the measurement system 100. The measurement area / measurement point display section displays the classification class to which each measurement area belongs, as well as the measurement area and measurement points calculated by the methods of embodiments 1 and 2. The parameter display section displays the set classification class parameters. The measurement value display section displays the measurement values. These displays allow the user to confirm that the classification class parameters automatically searched for according to the flowchart in FIG. 2 are likely to be accurate.

[0054] FIG. 15 shows an example of a GUI provided by the measurement system 100. In addition to the GUI described in FIG. 14, if the measurement area includes an outlier, a warning to that effect is displayed. The user can correct the measurement area in the measurement area correction unit. It may be possible to correct not only the measurement area warned as an outlier but also other measurement areas. Furthermore, it may be possible to instruct the measurement system 100 to re-learn the measurement area inference model D102 using the corrected measurement area. This makes it less likely that the measurement area inference model D102 will output an incorrect measurement area, making it possible to improve the performance of the measurement area inference unit 101 in a short period of time.

[0055] <Modifications of the Present Invention> The present invention is not limited to the above-described embodiments and includes various modifications. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and it is not necessary to include all of the configurations described. Furthermore, a part of one embodiment can be replaced with a configuration of another embodiment. Furthermore, a configuration of another embodiment can be added to a configuration of one embodiment. Furthermore, a part of the configuration of each embodiment can be added to, deleted from, or substituted for a part of the configuration of another embodiment.

[0056] In the above embodiment, the measurement processing unit 104 performs measurement using measurement parameters corresponding to the feature amounts of the measurement area. In addition to or instead of the measurement parameters corresponding to the feature amounts of the measurement area, measurement may be performed using a measurement method corresponding to the feature amounts of the measurement area. The measurement method here refers to an algorithm or the like that performs measurement processing on measurement points on an image.

[0057] In the above embodiment, each functional unit (measurement area inference unit 101, feature amount classification unit 102, parameter setting unit 103, measurement processing unit 104, and other functional units described in each figure) provided in the measurement system 100 can be configured by hardware such as a circuit device that implements these functions, or can be configured by a calculation device such as a CPU (Central Processing Unit) that executes software that implements these functions.

[0058] 100: Measurement system 101: Measurement area inference unit 102: Feature classification unit 103: Parameter setting unit 104: Measurement processing unit

Claims

1. A measurement system for measuring the structure of a sample, comprising: a measurement area inference unit that infers a measurement area to be measured in an image of the sample; and a measurement processing unit that performs measurement on the measurement area using measurement parameters corresponding to the features of the measurement area inferred by the measurement area inference unit, wherein the measurement area inference unit infers the measurement area from the image and extracts feature amounts of the inferred measurement area, and the measurement processing unit performs the measurement using the measurement parameters corresponding to the features, by using the measurement parameters corresponding to the feature amounts.

2. The measurement system described in claim 1, characterized in that the measurement area inference unit obtains a measurement area inference model that has been constructed in advance by machine learning so as to infer the measurement area from the image and extract features of the inferred measurement area, and the measurement area inference unit inputs the image into the measurement area inference model, thereby extracting the measurement area from the image and extracting features of the extracted measurement area.

3. The measurement system according to claim 2, further comprising a measurement area inference model update unit that updates the measurement area inference model, wherein the measurement area inference unit infers a training measurement area contained in the training image data by inputting the training image data into the measurement area inference model, the measurement area inference model update unit calculates the difference between the training measurement area contained in the training image data and the training measurement area inferred by the measurement area inference unit, and the measurement area inference model update unit updates the measurement area inference model so as to reduce the difference.

4. The measurement system according to claim 1, further comprising a feature classification unit that classifies the feature, and the measurement processing unit performs the measurement using the measurement parameters that correspond to the features of the measurement area by using the measurement parameters that correspond to the classification classes classified by the feature classification unit.

5. The measurement system according to claim 4, characterized in that the feature classification unit acquires a feature classification model configured to classify the feature into the classification class, and the feature classification unit classifies the feature into the classification class by inputting the feature into the feature classification model.

6. The measurement system described in claim 5, wherein the measurement area inference unit is configured to infer the measurement area from the image and extract features of the inferred measurement area using a measurement area inference model that has been constructed in advance by machine learning so as to infer the measurement area from the image and extract features of the inferred measurement area, and the measurement system further comprises a feature amount classification model learning unit that constructs the feature amount classification model, and the feature amount classification model learning unit constructs the feature amount classification model so as to classify features extracted from training data using the measurement area inference model.

7. The measurement system according to claim 1, further comprising a parameter setting unit that sets the measurement parameters, wherein the parameter setting unit sets the measurement parameters corresponding to the feature amounts, and the measurement processing unit uses the measurement parameters set by the parameter setting unit to perform the measurement using the measurement parameters corresponding to the features of the measurement area.

8. The measurement system according to claim 7, further comprising a measurement parameter search unit that searches for the measurement parameters, wherein the measurement processing unit performs measurement processing using image data for parameter search and a result of classifying the image data for parameter search based on feature amounts, wherein the measurement parameter search unit calculates an index value that represents the accuracy of the measurement processing for the image data for parameter search, and wherein the measurement parameter search unit acquires the measurement parameters used by the measurement processing unit by searching for the measurement parameters that improve the index value.

9. The measurement system according to claim 1, wherein the measurement processing unit performs the measurement by using a measurement method corresponding to the feature amount in addition to the measurement parameters corresponding to the feature amount.

10. The measurement system according to claim 1, further comprising an outlier reporting unit that detects outliers of the feature quantities, and the measurement processing unit uses the measurement parameters corresponding to the feature quantities excluding the outliers detected by the outlier reporting unit.

11. The measurement system according to claim 1, characterized in that the measurement system provides a user interface that presents the measurement results obtained by the measurement processing unit.

12. The measurement system according to claim 4, wherein the measurement system provides a user interface that presents class parameters that identify the classification classes.

13. The measurement system according to claim 10, wherein the measurement system provides a user interface that presents the measurement region corresponding to the outlier.

14. The measurement system described in claim 3, characterized in that the measurement system provides a user interface that allows a user to modify the measurement area inferred by the measurement area inference unit, and the measurement area inference model update unit updates the measurement area inference model using the modified measurement area.

Images